Research On Trademark Image Retrieval Technology Based On Multiple Feature Fusion

Trademarks are an important manifestation of goods and an important symbol used to distinguish goods or services between operators.Under the conditions of a market economy,in order to increase the visibility of their products and increase their sales,companies that produce and operate similar products will design a unique trademark for their own company to distinguish it from other companies.Therefore,strengthening the retrieval of trademark images can effectively prevent the occurrence of trademark fraud,cybersquatting and other phenomena,so as to maintain the exclusive right of enterprises to trademarks,protect the rights and interests of consumers,and ensure the legality and uniqueness of trademarks.According to the characteristics of the trademark image itself and the methods used in the current trademark image retrieval,we compared the trademark image with other types of images and found that the factors that determine the characteristics of the trademark image mainly include shape characteristics and color characteristics.Moreover,the trademark images of the same type have changes in scale,spatial position,etc.,so the retrieval completed by extracting a single feature or dual features is necessarily inaccurate.So how to accurately and quickly realize trademark image retrieval has become an urgent problem to be solved.Combined with the analysis of the above problems,this paper proposes an image retrieval algorithm based on multi-feature fusion.Our method is divided into three stages: the first stage uses the improved algorithm to extract the shape features,local scale invariant and color features of the trademark image;In the second stage,these three features are balanced by assigning different weights to obtain feature vectors that accurately describe the overall features of the trademark image;In the third stage,we use cosine similarity to perform similarity retrieval on the sample images to be retrieved and the verification sample.We have tested this method on a database containing more than 8,000 trademark images,and then compared the algorithm in this paper with the trademark image retrieval method of single feature and dual feature extraction by setting a comparison experiment.In addition,in order to accurately measure the performance of the algorithm,the PR curve is introduced as an evaluation criterion.Experimental results show that the algorithm mentioned in this paper is significantly better than other comparison algorithms in retrieval accuracy and retrieval time,which has good practical value in the field of well-known trademark detection and trademark copyright protection.